"strings"
"sync"
- "git.curoverse.com/arvados.git/sdk/go/arvados"
- "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
- "git.curoverse.com/arvados.git/sdk/go/auth"
- "git.curoverse.com/arvados.git/sdk/go/health"
- "git.curoverse.com/arvados.git/sdk/go/httpserver"
- "git.curoverse.com/arvados.git/sdk/go/keepclient"
- log "github.com/sirupsen/logrus"
+ "git.arvados.org/arvados.git/sdk/go/arvados"
+ "git.arvados.org/arvados.git/sdk/go/arvadosclient"
+ "git.arvados.org/arvados.git/sdk/go/auth"
+ "git.arvados.org/arvados.git/sdk/go/ctxlog"
+ "git.arvados.org/arvados.git/sdk/go/health"
+ "git.arvados.org/arvados.git/sdk/go/httpserver"
+ "git.arvados.org/arvados.git/sdk/go/keepclient"
+ "github.com/sirupsen/logrus"
"golang.org/x/net/webdav"
)
var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
+var notFoundMessage = "404 Not found\r\n\r\nThe requested path was not found, or you do not have permission to access it.\r"
+var unauthorizedMessage = "401 Unauthorized\r\n\r\nA valid Arvados token must be provided to access this resource.\r"
+
// parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
// PDH (even if it is a PDH with "+" replaced by " " or "-");
// otherwise "".
}
func (h *handler) setup() {
- h.clientPool = arvadosclient.MakeClientPool()
+ // Errors will be handled at the client pool.
+ arv, _ := arvados.NewClientFromConfig(h.Config.cluster)
+ h.clientPool = arvadosclient.MakeClientPoolWith(arv)
keepclient.RefreshServiceDiscoveryOnSIGHUP()
keepclient.DefaultBlockCache.MaxBlocks = h.Config.cluster.Collections.WebDAVCache.MaxBlockEntries
// are ignored (all response writes return the update error).
type updateOnSuccess struct {
httpserver.ResponseWriter
+ logger logrus.FieldLogger
update func() error
sentHeader bool
err error
if err, ok := uos.err.(*arvados.TransactionError); ok {
code = err.StatusCode
}
- log.Printf("update() changes response to HTTP %d: %T %q", code, uos.err, uos.err)
+ uos.logger.WithError(uos.err).Errorf("update() returned error type %T, changing response to HTTP %d", uos.err, code)
http.Error(uos.ResponseWriter, uos.err.Error(), code)
return
}
}
)
+func stripDefaultPort(host string) string {
+ // Will consider port 80 and port 443 to be the same vhost. I think that's fine.
+ u := &url.URL{Host: host}
+ if p := u.Port(); p == "80" || p == "443" {
+ return strings.ToLower(u.Hostname())
+ } else {
+ return strings.ToLower(host)
+ }
+}
+
// ServeHTTP implements http.Handler.
func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
h.setupOnce.Do(h.setup)
- var statusCode = 0
- var statusText string
-
- remoteAddr := r.RemoteAddr
- if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
- remoteAddr = xff + "," + remoteAddr
- }
if xfp := r.Header.Get("X-Forwarded-Proto"); xfp != "" && xfp != "http" {
r.URL.Scheme = xfp
}
w := httpserver.WrapResponseWriter(wOrig)
- defer func() {
- if statusCode == 0 {
- statusCode = w.WroteStatus()
- } else if w.WroteStatus() == 0 {
- w.WriteHeader(statusCode)
- } else if w.WroteStatus() != statusCode {
- log.WithField("RequestID", r.Header.Get("X-Request-Id")).Warn(
- fmt.Sprintf("Our status changed from %d to %d after we sent headers", w.WroteStatus(), statusCode))
- }
- if statusText == "" {
- statusText = http.StatusText(statusCode)
- }
- }()
if strings.HasPrefix(r.URL.Path, "/_health/") && r.Method == "GET" {
h.healthHandler.ServeHTTP(w, r)
if method := r.Header.Get("Access-Control-Request-Method"); method != "" && r.Method == "OPTIONS" {
if !browserMethod[method] && !webdavMethod[method] {
- statusCode = http.StatusMethodNotAllowed
+ w.WriteHeader(http.StatusMethodNotAllowed)
return
}
w.Header().Set("Access-Control-Allow-Headers", corsAllowHeadersHeader)
w.Header().Set("Access-Control-Allow-Methods", "COPY, DELETE, GET, LOCK, MKCOL, MOVE, OPTIONS, POST, PROPFIND, PROPPATCH, PUT, RMCOL, UNLOCK")
w.Header().Set("Access-Control-Allow-Origin", "*")
w.Header().Set("Access-Control-Max-Age", "86400")
- statusCode = http.StatusOK
return
}
if !browserMethod[r.Method] && !webdavMethod[r.Method] {
- statusCode, statusText = http.StatusMethodNotAllowed, r.Method
+ w.WriteHeader(http.StatusMethodNotAllowed)
return
}
w.Header().Set("Access-Control-Expose-Headers", "Content-Range")
}
+ if h.serveS3(w, r) {
+ return
+ }
+
pathParts := strings.Split(r.URL.Path[1:], "/")
var stripParts int
var attachment bool
var useSiteFS bool
credentialsOK := h.Config.cluster.Collections.TrustAllContent
+ reasonNotAcceptingCredentials := ""
- if r.Host != "" && r.Host == h.Config.cluster.Services.WebDAVDownload.ExternalURL.Host {
+ if r.Host != "" && stripDefaultPort(r.Host) == stripDefaultPort(h.Config.cluster.Services.WebDAVDownload.ExternalURL.Host) {
credentialsOK = true
attachment = true
} else if r.FormValue("disposition") == "attachment" {
attachment = true
}
+ if !credentialsOK {
+ reasonNotAcceptingCredentials = fmt.Sprintf("vhost %q does not specify a single collection ID or match Services.WebDAVDownload.ExternalURL %q, and Collections.TrustAllContent is false",
+ r.Host, h.Config.cluster.Services.WebDAVDownload.ExternalURL)
+ }
+
if collectionID = parseCollectionIDFromDNSName(r.Host); collectionID != "" {
// http://ID.collections.example/PATH...
credentialsOK = true
// data. Tokens provided with the request are
// ignored.
credentialsOK = false
+ reasonNotAcceptingCredentials = "the '/collections/UUID/PATH' form only works for public data"
}
}
if collectionID == "" && !useSiteFS {
- statusCode = http.StatusNotFound
+ http.Error(w, notFoundMessage, http.StatusNotFound)
return
}
}
formToken := r.FormValue("api_token")
- if formToken != "" && r.Header.Get("Origin") != "" && attachment && r.URL.Query().Get("api_token") == "" {
- // The client provided an explicit token in the POST
- // body. The Origin header indicates this *might* be
- // an AJAX request, in which case redirect-with-cookie
- // won't work: we should just serve the content in the
- // POST response. This is safe because:
+ origin := r.Header.Get("Origin")
+ cors := origin != "" && !strings.HasSuffix(origin, "://"+r.Host)
+ safeAjax := cors && (r.Method == http.MethodGet || r.Method == http.MethodHead)
+ safeAttachment := attachment && r.URL.Query().Get("api_token") == ""
+ if formToken == "" {
+ // No token to use or redact.
+ } else if safeAjax || safeAttachment {
+ // If this is a cross-origin request, the URL won't
+ // appear in the browser's address bar, so
+ // substituting a clipboard-safe URL is pointless.
+ // Redirect-with-cookie wouldn't work anyway, because
+ // it's not safe to allow third-party use of our
+ // cookie.
//
- // * We're supplying an attachment, not inline
- // content, so we don't need to convert the POST to
- // a GET and avoid the "really resubmit form?"
- // problem.
- //
- // * The token isn't embedded in the URL, so we don't
- // need to worry about bookmarks and copy/paste.
+ // If we're supplying an attachment, we don't need to
+ // convert POST to GET to avoid the "really resubmit
+ // form?" problem, so provided the token isn't
+ // embedded in the URL, there's no reason to do
+ // redirect-with-cookie in this case either.
reqTokens = append(reqTokens, formToken)
- } else if formToken != "" && browserMethod[r.Method] {
- // The client provided an explicit token in the query
- // string, or a form in POST body. We must put the
- // token in an HttpOnly cookie, and redirect to the
- // same URL with the query param redacted and method =
- // GET.
+ } else if browserMethod[r.Method] {
+ // If this is a page view, and the client provided a
+ // token via query string or POST body, we must put
+ // the token in an HttpOnly cookie, and redirect to an
+ // equivalent URL with the query param redacted and
+ // method = GET.
h.seeOtherWithCookie(w, r, "", credentialsOK)
return
}
}
if tokens == nil {
- tokens = append(reqTokens, h.Config.cluster.Users.AnonymousUserToken)
+ tokens = reqTokens
+ if h.Config.cluster.Users.AnonymousUserToken != "" {
+ tokens = append(tokens, h.Config.cluster.Users.AnonymousUserToken)
+ }
+ }
+
+ if tokens == nil {
+ if !credentialsOK {
+ http.Error(w, fmt.Sprintf("Authorization tokens are not accepted here: %v, and no anonymous user token is configured.", reasonNotAcceptingCredentials), http.StatusUnauthorized)
+ } else {
+ http.Error(w, fmt.Sprintf("No authorization token in request, and no anonymous user token is configured."), http.StatusUnauthorized)
+ }
+ return
}
if len(targetPath) > 0 && targetPath[0] == "_" {
arv := h.clientPool.Get()
if arv == nil {
- statusCode, statusText = http.StatusInternalServerError, "Pool failed: "+h.clientPool.Err().Error()
+ http.Error(w, "client pool error: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
return
}
defer h.clientPool.Put(arv)
}
}
// Something more serious is wrong
- statusCode, statusText = http.StatusInternalServerError, err.Error()
+ http.Error(w, "cache error: "+err.Error(), http.StatusInternalServerError)
return
}
if collection == nil {
// for additional credentials would just be
// confusing), or we don't even accept
// credentials at this path.
- statusCode = http.StatusNotFound
+ http.Error(w, notFoundMessage, http.StatusNotFound)
return
}
for _, t := range reqTokens {
if tokenResult[t] == 404 {
// The client provided valid token(s), but the
// collection was not found.
- statusCode = http.StatusNotFound
+ http.Error(w, notFoundMessage, http.StatusNotFound)
return
}
}
// data that has been deleted. Allow a referrer to
// provide this context somehow?
w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
- statusCode = http.StatusUnauthorized
+ http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
return
}
kc, err := keepclient.MakeKeepClient(arv)
if err != nil {
- statusCode, statusText = http.StatusInternalServerError, err.Error()
+ http.Error(w, "error setting up keep client: "+err.Error(), http.StatusInternalServerError)
return
}
kc.RequestID = r.Header.Get("X-Request-Id")
fs, err := collection.FileSystem(client, kc)
if err != nil {
- statusCode, statusText = http.StatusInternalServerError, err.Error()
+ http.Error(w, "error creating collection filesystem: "+err.Error(), http.StatusInternalServerError)
return
}
writefs, writeOK := fs.(arvados.CollectionFileSystem)
targetIsPDH := arvadosclient.PDHMatch(collectionID)
if (targetIsPDH || !writeOK) && writeMethod[r.Method] {
- statusCode, statusText = http.StatusMethodNotAllowed, errReadOnly.Error()
+ http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
return
}
// collection can't be saved.
w = &updateOnSuccess{
ResponseWriter: w,
+ logger: ctxlog.FromContext(r.Context()),
update: func() error {
return h.Config.Cache.Update(client, *collection, writefs)
}}
LockSystem: h.webdavLS,
Logger: func(_ *http.Request, err error) {
if err != nil {
- log.Printf("error from webdav handler: %q", err)
+ ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler")
}
},
}
}
openPath := "/" + strings.Join(targetPath, "/")
- if f, err := fs.Open(openPath); os.IsNotExist(err) {
+ f, err := fs.Open(openPath)
+ if os.IsNotExist(err) {
// Requested non-existent path
- statusCode = http.StatusNotFound
+ http.Error(w, notFoundMessage, http.StatusNotFound)
+ return
} else if err != nil {
// Some other (unexpected) error
- statusCode, statusText = http.StatusInternalServerError, err.Error()
- } else if stat, err := f.Stat(); err != nil {
+ http.Error(w, "open: "+err.Error(), http.StatusInternalServerError)
+ return
+ }
+ defer f.Close()
+ if stat, err := f.Stat(); err != nil {
// Can't get Size/IsDir (shouldn't happen with a collectionFS!)
- statusCode, statusText = http.StatusInternalServerError, err.Error()
+ http.Error(w, "stat: "+err.Error(), http.StatusInternalServerError)
} else if stat.IsDir() && !strings.HasSuffix(r.URL.Path, "/") {
// If client requests ".../dirname", redirect to
// ".../dirname/". This way, relative links in the
h.serveDirectory(w, r, collection.Name, fs, openPath, true)
} else {
http.ServeContent(w, r, basename, stat.ModTime(), f)
- if r.Header.Get("Range") == "" && int64(w.WroteBodyBytes()) != stat.Size() {
+ if wrote := int64(w.WroteBodyBytes()); wrote != stat.Size() && w.WroteStatus() == http.StatusOK {
// If we wrote fewer bytes than expected, it's
// too late to change the real response code
// or send an error message to the client, but
// at least we can try to put some useful
// debugging info in the logs.
n, err := f.Read(make([]byte, 1024))
- statusCode, statusText = http.StatusInternalServerError, fmt.Sprintf("f.Size()==%d but only wrote %d bytes; read(1024) returns %d, %s", stat.Size(), w.WroteBodyBytes(), n, err)
-
+ ctxlog.FromContext(r.Context()).Errorf("stat.Size()==%d but only wrote %d bytes; read(1024) returns %d, %v", stat.Size(), wrote, n, err)
}
}
}
+func (h *handler) getClients(reqID, token string) (arv *arvadosclient.ArvadosClient, kc *keepclient.KeepClient, client *arvados.Client, release func(), err error) {
+ arv = h.clientPool.Get()
+ if arv == nil {
+ err = h.clientPool.Err()
+ return
+ }
+ release = func() { h.clientPool.Put(arv) }
+ arv.ApiToken = token
+ kc, err = keepclient.MakeKeepClient(arv)
+ if err != nil {
+ release()
+ return
+ }
+ kc.RequestID = reqID
+ client = (&arvados.Client{
+ APIHost: arv.ApiServer,
+ AuthToken: arv.ApiToken,
+ Insecure: arv.ApiInsecure,
+ }).WithRequestID(reqID)
+ return
+}
+
func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []string, credentialsOK, attachment bool) {
if len(tokens) == 0 {
w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
- http.Error(w, http.StatusText(http.StatusUnauthorized), http.StatusUnauthorized)
+ http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
return
}
if writeMethod[r.Method] {
http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
return
}
- arv := h.clientPool.Get()
- if arv == nil {
- http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
- return
- }
- defer h.clientPool.Put(arv)
- arv.ApiToken = tokens[0]
-
- kc, err := keepclient.MakeKeepClient(arv)
+ fs, err := h.Config.Cache.GetSession(tokens[0])
if err != nil {
http.Error(w, err.Error(), http.StatusInternalServerError)
return
}
- kc.RequestID = r.Header.Get("X-Request-Id")
- client := (&arvados.Client{
- APIHost: arv.ApiServer,
- AuthToken: arv.ApiToken,
- Insecure: arv.ApiInsecure,
- }).WithRequestID(r.Header.Get("X-Request-Id"))
- fs := client.SiteFileSystem(kc)
+ fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
f, err := fs.Open(r.URL.Path)
if os.IsNotExist(err) {
http.Error(w, err.Error(), http.StatusNotFound)
LockSystem: h.webdavLS,
Logger: func(_ *http.Request, err error) {
if err != nil {
- log.Printf("error from webdav handler: %q", err)
+ ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler")
}
},
}
return nil
}
if err := walk(""); err != nil {
- http.Error(w, err.Error(), http.StatusInternalServerError)
+ http.Error(w, "error getting directory listing: "+err.Error(), http.StatusInternalServerError)
return
}
}
tmpl, err := template.New("dir").Funcs(funcs).Parse(dirListingTemplate)
if err != nil {
- http.Error(w, err.Error(), http.StatusInternalServerError)
+ http.Error(w, "error parsing template: "+err.Error(), http.StatusInternalServerError)
return
}
sort.Slice(files, func(i, j int) bool {
// into a cookie unless the current vhost
// (origin) serves only a single collection or
// we are in TrustAllContent mode.
- w.WriteHeader(http.StatusBadRequest)
+ http.Error(w, "cannot serve inline content at this URL (possible configuration error; see https://doc.arvados.org/install/install-keep-web.html#dns)", http.StatusBadRequest)
return
}
Value: auth.EncodeTokenCookie([]byte(formToken)),
Path: "/",
HttpOnly: true,
+ SameSite: http.SameSiteLaxMode,
})
}
if location != "" {
newu, err := u.Parse(location)
if err != nil {
- w.WriteHeader(http.StatusInternalServerError)
+ http.Error(w, "error resolving redirect target: "+err.Error(), http.StatusInternalServerError)
return
}
u = newu